Impact of Intelligent Speed Adaptation systems
on fuel consumption and driver behaviour
Guillaume Saint Pierre and Jacques Ehrlich
Abstract
In 1999, the French Ministry of Transport launched a significant program of experimentation and evaluation in order to assess the effects of different kinds of intelligent
speed adaptation systems in terms of acceptance by the drivers and effectiveness of speed
reduction in their daily trips. The LAVIA (Limiteur s’Adaptant à la VItesse Autorisée)
has been tested, in the Yvelines (France) according to three variants : advisory, voluntary
limited and mandatory limited. An experiment carried out over one year on a sample
of hundred drivers using twenty equipped vehicles, allowed recording a huge amount of
data in naturalistic driving conditions. Important results on safety were depicted in several
papers. Besides the initial objectives of the projects which were restricted to acceptance
and safety impact studies, it appears that a key aspect, in line with sustainable mobility
stakes, concerns LAVIA impact on fuel consumption. In this paper, after a will brief recall
of LAVIA objectives, technical aspects and results, we will present the results of a new
statistical analysis study which focuses on the respective impact of the three variants of
the system on fuel consumption.
I. INTRODUCTION
Speed management is a preoccupation for public authority as speed is considered as
the main cause of traffic injury accidents. Traditional methods of limiting speed have
only been moderately effective. Using the latest intelligent transportation technology,
speed enforcement can be enhanced through vehicle speed management programs,
often referred to as Intelligent Speed Adaptation (ISA). An ISA system monitors
the location and speed of the vehicle, compares it to a defined set speed, and takes
corrective action such as advising the driver and/or governing the top speed of the
vehicle. ISA systems effects on speed reduction and safety improvement is now
well known (see [3] [4] or [2]), and a brief recall of the LAVIA project results
on this aspect is given in this paper. Moreover, speed is closely related to fuel
consumption and CO2 emissions rates, and many previous research projects on
the usage of ISA systems are predicting a significant fuel consumption reduction.
As fuel consumption is hard to measure, previous studies make use of state-ofthe-art transportation/emissions modeling tools. This paper address the problem by
using an instantaneous fuel consumption estimation model based on precise engine
knowledge made available by OEM manufacturers partners of the LAVIA project.
This work was supported by the French National Research Agency
G. Saint Pierre is with Vehicle-Infrastructure-Driver Interactions Research Unit, Joint unit INRETS-LCPC,
Bat 824, 14 route de la minière, 78000 Versailles-Satory, France [email protected]
J. Ehrlich is the head of the Vehicle-Infrastructure-Driver Interactions Research Unit,
[email protected]
There are many forms of ISA, most of them relying on modern technology such as
Global Position System (GPS) receivers, on-board roadway databases, and/or wireless
communication. There are several ISA implementation methods, based on how the
set speed is determined [2] :
1) fixed : in this case, the maximum permissible speed is set by the user and
the on-board control system never exceeds that value ; for this, ISA can be
implemented as an independent on-board control system.
2) variable : in this case, the set speed is determined by vehicle location, where
different speed limits are set spatially. This is the most common implementation
of ISA, where the maximum vehicle speed never exceeds the speed limit for a
given area. This can be implemented based solely on position information or
based on broadcasted values. LAVIA system is one of them.
3) dynamic : in this case, speed is determined by time and location. The temporal
aspect can vary based on road network conditions or weather. This information can be provided from a transportation management center via vehicleinfrastructure communication.
Another dimension to ISA systems is how it intervenes with driver behavior. Categories include :
1) advisory, where limits are displayed on a messaging device and the driver
changes vehicle speed accordingly ;
2) active support, where the control system can change vehicle speed but driver
can override ; and
3) mandatory, where ISA controls maximum speed and driver cannot override.
Concerning the variable ISA implementation, the most frequently cited result about
possible fuel consumption reduction using ISA systems is a conclusion of an ambitious research experimentation that took place in England and made by the University
of Leeds ([2]). Using simulation methods and with the assumption of a full ISA
penetration, [1] claim ISA leads to a reduction in fuel consumption of about 8%
in urban areas, whereas only 2% reduction is expected on highways. Other studies
like [4] using also simulation models estimate the fuel consumption reduction around
11%. Those results leads public authorities to expect wide benefits from ISA systems,
but some recent studies tend to mitigate those hopes. For example, the ISA evaluation
made recently by the Monash University (Australia, see [6]) do not highlight any
fuel reduction due to usage of ISA, except for 80km/hr limites zones, and for the
combination of ISA and following distance warning (FDW) systems only. Concerning
dynamic ISA systems, [8] are more positive. Using both simulations and real-world
experiments, mainly during high levels of congestion, they find a 13% percent fuel
consumption reduction.
It is of great interest to investigate the reasons of such different impact on fuel
consumption. This paper address this assumption by means of statistical analysis
based on data recorded in naturalistic driving conditions during the LAVIA experimentation and fuel consumption experimental model from a French car manufacturer
which was partner of the LAVIA project.
II. T HE LAVIA PROGRAM AND EXPERIMENTATION
The Directorate for traffic and road safety of the French Ministry of Transport commissioned the National Research Institute on Transportation and Safety plus others
public partners to investigate and to assess the potential of the LAVIA for helping
the drivers to respect the speed limits by means of an on-road experimentation. Three
man-machine interaction modes of the LAVIA system have to be tested : an advisory
mode and two active modes (voluntary and mandatory). The project was conducted
in three phases. Phase 1 involved the development of functional specifications for
the LAVIA system and the data recorder system. In phase 2, the pre-evaluation
phase, two prototypes have been driven over a fixed route of 90 kilometres by a
set of 20 voluntary drivers to test for acceptance and usage against the original
specifications. Phase 3, the naturalisitic driving evaluation, was the experimental
phase conducted on a sample of one hundred drivers during one year, on wide area
with different kind of road networks : urban, rural and motorway. Each driver goes
successively through four modes of the LAVIA system by periods of two weeks :
neutral/ advisory/voluntary limited/mandatory limited. The different operating modes
were tested sequentially after a neutral mode (neither advisory nor active) in order
to acquire information about the normal driver’s behaviour as a reference for further
comparison. The experimental design is a within subject design, where the control
condition is the neutral mode. Usually that kind of experimental design needs to
be counterbalanced. This means that one half of the subjects should start with the
control condition and then switch to the experimental condition (active mode), and
half of the subjects should do vice versa. Due to practical reasons, this has not
been the case in the LAVIA experiment and there is a risk of the presence of a
learning effect. LAVIA vehicles were equipped with a data acquisition system which
recorded every 500 ms numerous driving parameters such as position, actual speed,
speed limit, driver’s action on kick-down, gas pedal, brake, windshield wiper etc.
Thus a huge amount of data has been made available for different kind of statistical
analysis such as usage, utility, usability and safety impact.
III. F UEL CONSUMPTION DATA COLLECTION
As the LAVIA project was not originally designed to include consumption data
collection, a specific approach has been carried out after the overall experimentation
using both part of the previous collected data and precise knowledge of the engine
behaviour. Thanks to an experimental fuel consumption model, it became possible
to relate numerically the pressure on the accelerator pedal and the engine rpm to
the volume of fuel injected each 500ms interval. This was possible for half of the
involved vehicle fleet, leading to a data set of 44 drivers observed during 8 weeks and
using the four different modes of the LAVIA system. This experimental model makes
useless the usage of classical models for the estimation of fuel consumption that had
wide spread use by many researchers (like the Positive Kinetic Energy model).
IV. E ARLIER STATISTICAL CONCLUSIONS
In the first phase of the statistical analysis process, attention was focused on speed
reduction and acceptability of the system itself. Consideration has been given to the
sample of trips (frequency, average length or duration, proportion by trip motivation)
to ensure a valid comparability between modes. In [3], speed distributions have been
estimated showing clearly behaviours changes especially in active mode. The results
showed that compliance with the speed limits improved among test drivers, leading
to a peak in speed distribution around the limitation. As a potential drawback of the
system, drivers tend to adopt speeds very close to the limitation as the accelerator
pedal is controlled by the system in the case of a speed violation. The conclusions
show that all active modes had positive effects on safety, with the largest effect of
the mandatory system. Besides the safety improvement, it is often claimed that ISA
also leads to a reduction in fuel consumption. This work address this assumption by
means of multivariate statistical analysis based on indicators computed from each
trip data.
V. M ETHODOLOGY
First approach developed in this work is based on the observed trips. With regard to
vehicle variables (speed, accelerations, engine rpm, pressure on the accelerator pedal
etc.), a range of indicators exist that could potentially yield some quantification as
to the level of unsteadiness present in the traffic stream. In the same way, various
statistical indicators can be related to the traffic conditions (percentage of vehicle
stationary time, infrastructure type etc.) and the driver behaviour (percentage of time
above the speed limit, gearshift strategy, acceleration peaks, brakes usage etc.). Such
indicators are aggregated across all the six thousand trips made by the 44 studied
drivers, and analyzed using multivariate statistical methods. Very short trips are
excluded and 200 seconds at the beginning and the end of the trip are censored.
The relationship between qualitative variables such as the mode of the LAVIA
(active or not) or the trip motivation and quantitative variables such as average fuel
consumption are highlighted using multiway anova and correspondence analysis.
Clustering methods associated with multivariate analysis methods (PCA) are also
useful to obtain homogeneous groups of trips in term of traffic conditions.
VI. DATA ANALYSIS
Field trials studies show positive effects on speed behaviour and expect significant
effects on fuel consumption reduction. To our knowledge, this has just been addressed
using microsimulation modeling and fuel reduction induced by the use of ISA
systems is still to prove in a large field trial. We claim that the LAVIA system do
not generally induce significant fuel savings. Three types of questions are currently
under investigation :
– How drivers use the LAVIA system in its limited mode ?
– What is the impact on the driver behaviour concerning gear choice in relation
with speed ?
Average consumption
Neutral
Advisory
11.4
11.7
Voluntary
limited
12.2
Mandatory
limited
11.6
F
p-value
6.21
0.0003
TABLE I
A NALYSIS OF VARIANCE FOR THE AVERAGE FUEL CONSUMPTION UNDER THE FOUR VARIANTS OF THE
LAVIA SYSTEM .
– What is the effect of three modes of LAVIA on fuel consumption ?
Answering those questions is not a simple task because of the complexity and the
variability of the driving situations. As described earlier, we construct a data set where
the trips are considered as the statistical individuals. Those trips can be stratified with
respect to some carefully chosen variables, but the first approach we present take
account of the entire journey. The statistical analysis methods required to explore such
a data set belong to the data mining area. This analyse will be described in the first
subsection, while stratified approaches are conducted in the following subsections.
A. Exploration of the dataset
As usual, data mining approaches need a multidimensional exploration of the data
set. In order to simplify the data set, this first approach make use of quantitative
variables transformed to qualitative ones. Some correspondence analysis have been
done (not showed here) using those transformed variables, leading to the conclusion
that the major part of the variability of the average fuel consumption is explained by
classical factors. The great influence of the percentage of time with the vehicle
stopped is clear. This indicator is related to the congestion, and it is clear for
everybody that congestion induce higher fuel consumption. This indicator is also
related to the infrastructure type : congestions are more likely to occur in urban
areas. As congestions induce frequent gearshifts, we also observe that the more the
driver shift the gears, with respect to the time, the more the average consumption
is high. This is clear if we consider the fact that fuel consumption is not so high
on highways. This is due to the constant speed at a low engine rpm and without
gearshifts. Taking account of the entire data set without any precautions leads to
possible biased analysis. It is difficult to extract the information related to the usage
of the LAVIA system because of the great variability inducted by variables such
the presence of congestions, infrastructure type or speed limit value. First analysis
conducted is presented at table V. This analysis of variance (anova) show a strong
difference between the average fuel consumption under the four tested conditions
(p-value<0.05), with higher values for the voluntary limited variant. Looking more
deeply at this surprising phenomenon with the help of a multiple comparison test
show that the difference is due to the voluntary limited variant only. This mean
that there is no significant difference between Neutral/Advisory/Mandatory limited
modes. As already highlighted, the tested variant order in the experiment is the same
25
for all subjects. Searching for the evidence of a learning effect, we plot boxplots of
the trip’s average fuel consumption related to the day. Recall that the neutral mode has
been tested the first 14 days, the advisory mode the following 14 days, and the last 28
days are used to test first the voluntary limited mode, and after the mandatory limited
mode. Looking at figure 1 we see that the average fuel consumption repartition follow
a decreasing trend, particularly if we consider the 95th percentile and the number of
outliers. Therefore, the significant difference between the 4 variants of the LAVIA
seems to be related to the learning effect at the beginning of the experiment with
the active system.
●
95th percentile
of observed mean fuel consumption
● ●
●
●
● ●
20
●
● ●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
15
●
●
●
●
●
●
●
●
●
●
●
●
● ●
●
10
●
●
5
Fuel consumption (liters per 100km)
●
●
●
●
●
●
●
●
30 32 34 36 38 40 42 44 46 48 50 52 54 56 58
Ordered experience days
Fig. 1. Boxplots for the trip’s mean consumption by the day of the experience. The line represent the 95th
percentile.
B. Stratified analysis
Data collected during the LAVIA project can be assimilated to data collected in
naturalistic driving conditions with many different observed driving conditions. The
influence of the context (trip motivation, speed limit value, infrastructure characteristics, traffic condition etc.) has to be evaluated and controlled to validate the
comparison of effects between modes. Fuel consumption is highly dependant on
the traffic conditions, and higher values are therefore associated with trips under
congested flow conditions characterized by small average speed, great percentage
of vehicle stationary time and a high gear-shifting rate. We need to be very careful
when controlling these traffic conditions to get proper and significant results. For
that we need to stratify the sample according to some carefully chosen variables in
order to deal with sections of comparable routes. The overarching principle of our
approach is to be reduced to comparable situations, using the controllable factors at
our disposal. Below are the various approaches proposed in this report.
1) classification algorithms : Study similar trips in terms of congestion, duration,
average speed and so on.
2) Common sense : Use the variable "goal of the travel" and study trips from
home to work.
3) Stratification according to the type of network.
4) Stratification according to the speed limits.
1) Driving conditions: The congestion phenomenon is crucial to deal with. Usually
this is done by analyzing only observations where the time headway between the
observed car and the vehicle ahead is greater than or equal to 3 seconds. This ensure
that the observed speeds are not constrained by the vehicle ahead. Unfortunately,
this variable was not collected during the LAVIA experiment. We use instead the
trip’s percentage of time with the vehicle stopped (speed=0), and we take account
of the trips with less than 20% of the time with a null speed. Before trying some
classification methods, we need to be careful in the variables choice. Classification
trees are used to choose the significant variables, leading to a subset of variables with
a great explanatory power. Variables used are the following : trip’s % of time with
speed equal to zero, % of time on urban areas, rural areas, and highways, average
speed, duration, mean fuel consumption, number of gearshifts over duration, % of
time above the speed limit, and the 95th percentile of the acceleration. Considering
the necessity of dimensionality reduction, we apply first a Principal Component
Analysis (PCA) before applying classification methods to the resulting subset. This
approach leads to identify 4 homogeneous groups in term of traffic conditions, more
tractable to highlight the LAVIA impact on fuel consumption.
The 4 groups found are : – Group 1 represents short-distance trips made in a urban
area with high fuel consumption. – Group 2 is the one with the largest proportions
of rural and highway infrastructure, with low fuel consumption and high average
speed. – Group 3 is the largest one, associated with a classical behaviour, both in
urban and rural areas. – Group 4 is the one with high average speeds, often beyond
the speed limits and mainly in a rural environment.
The result of the anova is presented in table II. It puts forward a statistically
significant influence of LAVIA system for groups 1, 2 and 3. The influence is very
clear for groups 1 and 3, and a little less for Group 2. As has been previously
discussed, there is a learning effect that leads to an increase in fuel consumption when
using voluntary limited mode. This is reflected here since a multiple comparisons
method only detect the influence of the voluntary limited mode. For the 4 groups
found, there is no difference between Neutral/Advisory/Mandatory limited modes.
There is actually a return to normal during the transition to the Mandatory limited
mode. The only group for which the voluntary limited mode do not appear to increase
consumption is group 4, where the average speed is very high and with great speed
overruns. Note that the significance of the results is bigger (higher F statistics) for
groups containing journeys made mainly in urban areas.
Group
Group
Group
Group
1
2
3
4
Neutral
Advisory
11.09
8.17
9.44
7.96
11.11
7.92
9.38
8.34
Voluntary
limited
11.83
8.58
9.91
8.57
Mandatory
limited
11.39
8.45
9.71
8.26
F
p-value
7.27
4.12
8.037
0.83
7.7e-5
0.006533
2.57e-05
0.48
TABLE II
S TRATIFICATION BY GROUPS OF SIMILAR DRIVING CONDITIONS . A NALYSIS OF VARIANCE FOR THE
AVERAGE FUEL CONSUMPTION UNDER THE FOUR VARIANTS OF THE
LAVIA SYSTEM .
2) Speed limits: Stratified the studied routes by the legal speed limit is another
way to obtain homogeneous and comparable data. On each portion of the journey
obtained after stratification is calculated the average consumption in liters/100km.
The anova for each speed limit is shown at table III. It is striking to see how the
impact of LAVIA system in active mode is dependent on the speed limit. Indeed,
the p-values are very low for 30km/h areas, then higher as the evolution of legal
speeds. It is very important to note again that only the voluntary limited mode causes
increased fuel consumption (methods of multiple comparisons). The LAVIA system
effect on fuel consumption therefore seems more important in areas where the legal
speed is low. Note also that the variability in average consumptions is always lower
when using LAVIA active modes (both voluntary or mandatory limited).
3) Infrastructure type: In order to stratify according to the infrastructure type, we
proceed in the same way as for the legal speeds. From what we can read from table
IV, only the urban network shows a significant difference in terms of consumption.
Zone
30km/h
Trip number
de trajets
1898
50km/h
6830
70km/h
2660
90km/h
2680
110km/h
1870
Fuel consumption
(litres/100km)
Mean
Standard deviation
Mean
Standard deviation
Mean
Standard deviation
Mean
Standard deviation
Mean
Standard deviation
Neutral
Advisory
16.79
5.17
13.59
3.72
9.02
3.01
8.57
3.23
7.58
1.95
17.24
5.11
13.62
3.97
9.21
3.09
8.75
3.44
7.66
2.31
Voluntary
limited
19.07
4.67
14.13
3.99
9.58
3.11
9.06
3.48
7.72
2.02
Mandatory
limited
18.29
4.85
13.61
3.4
9.23
2.84
8.83
3.15
7.50
1.93
F
p-value
20.91
<.0001
7.92
<.0001
4.0
0.0075
2.48
0.0592
0.98
0.4013
TABLE III
S TRATIFICATION BY THE LEGAL SPEED LIMIT. A NALYSIS OF VARIANCE FOR THE AVERAGE FUEL
CONSUMPTION UNDER THE FOUR VARIANTS OF THE
LAVIA SYSTEM .
Network
Urban
Rural
Highway
Fuel consumption
(litres/100km)
Mean
Mean
Mean
Neutral
Advisory
13.25
9.25
7.26
13.39
8.86
7.43
Voluntary
limited
13.93
9.42
7.71
Mandatory
limited
13.34
9.36
7.37
F
6.7067
2.3677
1.3926
p-value
0.0001626
0.06883
0.243
TABLE IV
S TRATIFICATION BY THE INFRASTRUCTURE TYPE . A NALYSIS OF VARIANCE FOR THE AVERAGE FUEL
CONSUMPTION UNDER THE FOUR VARIANTS OF THE
Inter-urbain
Urbain
Autoroute
Neutral
Advisory
9.07
13.2
7.5
9.36
13.4
7.8
Voluntary
limited
9.96
13.7
7.9
LAVIA SYSTEM .
Mandatory
limited
9.52
13.2
7.4
F
p-value
3.23
1.77
0.56
0.0217
0.1505
0.6386
TABLE V
S TRATIFICATION BY THE INFRASTRUCTURE TYPE , FOR THE TRIPS FROM HOUSE TO WORK ONLY ( AND
RESPECTIVELY, FROM WORK TO HOUSE ).
A NALYSIS OF VARIANCE FOR THE AVERAGE FUEL CONSUMPTION
UNDER THE FOUR VARIANTS OF THE
LAVIA SYSTEM .
But the difference comes again from the voluntary limited mode. This analysis is
therefore perfectly compatible with the previous one.
4) Objective of the journey: The last stratification considered in this work relates
to the trips between home and work. We know that they are the same for each
driver, even if it is not always the case due to data collection errors. Applying
an outliers detection method help to solve this problem. Indeed, the impact of a
driving assistance system as LAVIA is likely to be better seen for identical routes
between home and work (and respectively). Studying for each driver the change in
fuel consumption for different LAVIA modes on the route between home and work
is a way to bring an experimental type of framework allowing better controlling
of the external factors not collected (weather, holidays, strikes and so on.). Results
are displayed in table V. No significant difference in fuel consumption is observed
in urban areas and on highways. A low significant difference (p-value slightly less
than 0.05) is observed in rural areas, but a multiple comparison test only detects a
difference between neutral and voluntary limited mode. This difference is therefore
not very interesting and probably due to a bias related to the learning effect. Bias,
which has been observed several times in previous analyses.
5) Preliminary conclusion: The stratified analyses presented in this chapter agree
remarkably well on the following points. – Only the voluntary limited mode induces
significant differences in fuel consumption for all analyses. This helps to confirm the
existence of learning effect. – The existence of this learning effect implies that the
LAVIA system may have a negative impact on fuel consumption when it is poorly
controlled. – The impact of a bad usage of the LAVIA system is greater in urban
areas, ie in areas limited to 30km/h or 50km/h. – The mandatory limited mode allows
a return to a fuel consumption identical to the condition of control (Neutral mode).
The LAVIA system is a complex tool, which drivers have several days to tame.
When it is misused, this system leads to increases in fuel consumption. The next
section is devoted to some attempts to explain this phenomenon. Those LAVIA
drawbacks must be understood in order to avoid them. When the system is used
properly, after 2 weeks learning, there are no more increase in fuel consumption.
But we do not observe the promised fuel consumption reduction. The LAVIA system
therefore provides gains very clear in terms of security, but not in terms of fuel
consumption.
VII. I MPACT OF THE LAVIA SYSTEM ON THE DRIVING BEHAVIOUR
Subjects using LAVIA drives is a little slower, but it does not feel on fuel consumption. It appears that the driving activity is influenced by the LAVIA system, and that
strategies used by drivers are different.
A. Acceleration
The first phenomenon concern the pression on the accelerator pedal. We calculated
the variable giving the percentage of travel time for which the accelerator pedal is
pressed for more than 50% and 75 %. The graphs 2 show the boxplots of these
variables under the four LAVIA modes. Clearly, the LAVIA in its active modes
induces a pressure on the pedal much harder than for Neutral and advisory modes.
These overloads may be linked to kick-down, but given that this is a temporary
action, it seems more appropriate to seek another cause for this phenomenon. The
drivers knows that the injection is controlled by system, so they drive up to the
limit, without taking into account the pressure on the accelerator pedal. The impact
in terms of consumption is more debatable, but the cause could be the gear choice.
Indeed, the injection of gasoline is limited when the limitation is active, but the gear
is still defined by the user.
B. gearshift strategy
We are now entering a stratified analysis according to speed limits. It takes the
form of three types of graphs. The first (3) gives the observed gear choices for the
speed limit in question and for the 4 LAVIA variants. The second type of graph (4)
shows the repartition of the couples (Instant Speed ΠEngine rpm) for neutral and
mandatory limited modes. It was obtained using a Gaussian kernel density estimation
(KDE) in order to highlight the accumulation points (associated with peaks) separated
by lines levels.
In the case of the areas limited to 30km/h, Figure 3 shows that the gear distribution
is different between modes : active modes of LAVIA induce lower gear choices.
Looking at the graphs (4), we see that the space distribution of the couples (Speed
Instant ΠRegime Engine) is also different. The mandatory limited mode implies
●
●
●
●
30
●
●
●
●
●
●
●
●
●
●
25
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
20
●
●
15
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
5
10
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
0
% of time with the accelerator pedal pushed up over 50%
●
Advisory
Mandatory limited
Neutral
Voluntary limited
LAVIA variant
15
●
10
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
5
0
% of time with the accelerator pedal pushed up over 75%
(a) Accelerator pedal pushed-in over 50%.
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
Advisory
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
●
Mandatory limited
Neutral
Voluntary limited
LAVIA variant
(b) Accelerator pedal pushed-in over 75%.
Fig. 2.
Percentage of the trips with the accelerator pedal pushed-in over 50 and 75%.
that these couples do not go beyond 30km/h, but the choice of the second speed (2nd
cloud from left) is by far the most represented. In contrast, when using the Neutral
mode the gear choices are better distributed and the third speed is less adopted.
The problem is that the second speed is selected to drive at 30km/h with an
engine rpm of 2000tr/min, which is greater than the engine rpm associated with the
third gear. It is as if drivers using mandatory limited mode forgot how to switch to
the third gear. The result is an higher fuel consumption when travelling in second
gear at 30km/h (2000tr/min regime), that when we ride in the third 30km/h (regime
1500tr/min). This is confirmed by the fact that the probability of moving from the
second speed in the third speed mode neutral east of 20.1 % and only 17.2 % in
active mode.
100
40
60
80
Mandatory limited
Voluntary limited
Advisory
Neutral
29.1 29.9
20
25.8 25.9
17.8
14.2 14.1 15.1
2.5 1.5 2.2 2.3
0.6 0.4 0.5 0.7
Gear 4
Gear 5
0
12 12.8 12.4 11.5
Gear 1
Fig. 3.
Gear 2
Gear 3
Gear repartition for each variant of the LAVIA. Speed limit = 30km/h.
VIII. C ONCLUSION
Speed management is a major aspect of road transport, and a 100% observance of
the speed limits is often viewed as the perfect solution to a various set of problems
(safety, congestions, greenhouse emissions etc.). As an example, the estimate of the
Dutch Ministry of Transport of a 100% observance of the speed limits on road safety
(see [4]) is that road fatalities will be reduced by 20% and hospitalised wounded by
15%. Fuel consumption and carbon dioxide emission will be reduced by 11% each.
Such conclusions may be plausible considering the security aspects, but seems to be
excessive if we consider the fuel consumption aspects. As a system that guarantee
a strict observance of the speed limits, ISA system such the LAVIA should have
highlight a fuel consumption reduction. Instead of a reduction,
First results show that voluntary limited mode of the ISA system leads to a
significant increase in average fuel consumption, with less significance for the trips in
inter-urban and motorway areas. Moreover, stratifying the trips by legal speed values
show that only 30km/h zones (8% of the experimental zone) increase significantly
the fuel consumption whereas no significant difference were found on the other
zones. Investigations about possible explanations of this result lead to study precisely
driver’s behaviour and their gearshift strategies. Gear choices under active modes of
the LAVIA system are slightly different from those corresponding to inactive modes
respecting to the instant vehicle speed. It is therefore shown that active modes of the
LAVIA system induce diminished attention, especially in the gear choice.
IX. N EXT STEPS
Other analysis methods will be carried out such as spatiotemporal analysis using
GIS and digital maps with topology. More precise fuel consumption comparisons will
5000
5000
4000
4000
Density
Density
4e−05
3e−05
2e−05
1e−05
2000
1e−04
Engine_rpm
Engine_rpm
5e−05
3000
3000
8e−05
6e−05
4e−05
2e−05
2000
0e+00
0e+00
1000
1000
0
0
0
20
40
60
80
100
120
140
0
Instantaneous_Speed
20
40
60
80
100
120
140
Instantaneous_Speed
(a) Neutral mode, Speed limit = 30km/h
(b) Mandatory limited mode, Speed limit =
30km/h
Fig. 4. Non parametric kernel density estimation of the repartition of the couples (Instantaneous speed ×
Engine rpm).
be achieved by stratifying according to the routes instead the type of infrastructure.
Specific attention will be paid to characterize the driving behaviour under active
modes, and comparison with green-driving will be presented. Impact of the LAVIA
on the driver behaviour in term of green-driving will be achieved.
R EFERENCES
[1] O. Carsten, “Prosper results : Benefits and costs.” Presentation at the PROSPER Seminar on 23 November
2005 in Brussels., Tech. Rep., 2005.
[2] O. Carsten and M. Fowkes, “External vehicle speed control,” executive summary of project results. Institute
for Transport Studies, University of Leeds, Tech. Rep., 2000.
[3] J. Ehrlich, F. Saad, S. Lassarre, and S. Romon, “Assessment of lavia systems : experimental design and first
results on system use and speed behaviour,” Proceedings of the ITS World Congress, 2006.
[4] Ministry of Transport, “Field experiment intelligent speed adaptation,” ISA Tilburg. Final Report. Rotterdam,
[In Dutch]., Tech. Rep., 2001.
[5] R Development Core Team, R : A language and environment for statistical computing, R Development
Core Team, Vienna, Austria, 2005, ISBN 3-900051-07-0. [Online]. Available : http://www.R-project.org
[6] M. A. Regan, T. J. Triggs, K. L. Young, N. Tomasevic, E. Mitsopoulos, E. Stephan, and C. Tingvall, “Onroad evaluation of intelligent speed adaptation, following distance warning and seatbelt reminder systems :
Final results of the tac safecar project.” Monash University, Tech. Rep., 2006.
[7] SAS Institute, Base SAS 9.1 Procedures Guide. SAS Publishing, 2004.
[8] O. Servin, K. Boriboonsomsin, and M. Barth, “An energy and emissions impact evaluation of intelligent
speed adaptation.” In Proceedings of the IEEE ITSC 2006, pp. 1257–1262, 2006.